Bionic: x86: Fix libm macro definitions

http://code.google.com/p/android/issues/detail?id=19276

GCC would remove inline asm due to lack of knowledge
of FPU register changes.

Change-Id: I9f9e8623fa6580843b7cd8178439ace8c2db2d51
Signed-off-by: Mark D Horn <mark.d.horn@intel.com>
Signed-off-by: Bruce Beare <bruce.j.beare@intel.com>
Author: Jingwei Zhang <jingwei.zhang@intel.com>
diff --git a/libm/include/i387/fenv.h b/libm/include/i387/fenv.h
index 4281f10..710494c 100644
--- a/libm/include/i387/fenv.h
+++ b/libm/include/i387/fenv.h
@@ -45,13 +45,6 @@
 	char		__other[16];
 } fenv_t;
 
-#define	__get_mxcsr(env)	(((env).__mxcsr_hi << 16) |	\
-				 ((env).__mxcsr_lo))
-#define	__set_mxcsr(env, x)	do {				\
-	(env).__mxcsr_hi = (__uint32_t)(x) >> 16;		\
-	(env).__mxcsr_lo = (__uint16_t)(x);			\
-} while (0)
-
 typedef	__uint16_t	fexcept_t;
 
 /* Exception flags */
@@ -72,167 +65,35 @@
 #define	_ROUND_MASK	(FE_TONEAREST | FE_DOWNWARD | \
 			 FE_UPWARD | FE_TOWARDZERO)
 
-/*
- * As compared to the x87 control word, the SSE unit's control word
- * has the rounding control bits offset by 3 and the exception mask
- * bits offset by 7.
- */
-#define	_SSE_ROUND_SHIFT	3
-#define	_SSE_EMASK_SHIFT	7
-
-/* After testing for SSE support once, we cache the result in __has_sse. */
-enum __sse_support { __SSE_YES, __SSE_NO, __SSE_UNK };
-extern enum __sse_support __has_sse;
-int __test_sse(void);
-#ifdef __SSE__
-#define	__HAS_SSE()	1
-#else
-#define	__HAS_SSE()	(__has_sse == __SSE_YES ||			\
-			 (__has_sse == __SSE_UNK && __test_sse()))
-#endif
-
 __BEGIN_DECLS
 
 /* Default floating-point environment */
 extern const fenv_t	__fe_dfl_env;
 #define	FE_DFL_ENV	(&__fe_dfl_env)
 
-#define	__fldcw(__cw)		__asm __volatile("fldcw %0" : : "m" (__cw))
-#define	__fldenv(__env)		__asm __volatile("fldenv %0" : : "m" (__env))
-#define	__fnclex()		__asm __volatile("fnclex")
-#define	__fnstenv(__env)	__asm __volatile("fnstenv %0" : "=m" (*(__env)))
-#define	__fnstcw(__cw)		__asm __volatile("fnstcw %0" : "=m" (*(__cw)))
-#define	__fnstsw(__sw)		__asm __volatile("fnstsw %0" : "=a" (*(__sw)))
-#define	__fwait()		__asm __volatile("fwait")
-#define	__ldmxcsr(__csr)	__asm __volatile("ldmxcsr %0" : : "m" (__csr))
-#define	__stmxcsr(__csr)	__asm __volatile("stmxcsr %0" : "=m" (*(__csr)))
+/* C99 floating-point exception functions */
+int feclearexcept(int excepts);
+int fegetexceptflag(fexcept_t *flagp, int excepts);
+int fesetexceptflag(const fexcept_t *flagp, int excepts);
+/* feraiseexcept does not set the inexact flag on overflow/underflow */
+int feraiseexcept(int excepts);
+int fetestexcept(int excepts);
 
-static __inline int
-feclearexcept(int __excepts)
-{
-	fenv_t __env;
-	int __mxcsr;
+/* C99 rounding control functions */
+int fegetround(void);
+int fesetround(int round);
 
-	if (__excepts == FE_ALL_EXCEPT) {
-		__fnclex();
-	} else {
-		__fnstenv(&__env);
-		__env.__status &= ~__excepts;
-		__fldenv(__env);
-	}
-	if (__HAS_SSE()) {
-		__stmxcsr(&__mxcsr);
-		__mxcsr &= ~__excepts;
-		__ldmxcsr(__mxcsr);
-	}
-	return (0);
-}
-
-static __inline int
-fegetexceptflag(fexcept_t *__flagp, int __excepts)
-{
-	int __mxcsr, __status;
-
-	__fnstsw(&__status);
-	if (__HAS_SSE())
-		__stmxcsr(&__mxcsr);
-	else
-		__mxcsr = 0;
-	*__flagp = (__mxcsr | __status) & __excepts;
-	return (0);
-}
-
-int fesetexceptflag(const fexcept_t *__flagp, int __excepts);
-int feraiseexcept(int __excepts);
-
-static __inline int
-fetestexcept(int __excepts)
-{
-	int __mxcsr;
-	short __status;
-
-	__fnstsw(&__status);
-	if (__HAS_SSE())
-		__stmxcsr(&__mxcsr);
-	else
-		__mxcsr = 0;
-	return ((__status | __mxcsr) & __excepts);
-}
-
-static __inline int
-fegetround(void)
-{
-	int __control;
-
-	/*
-	 * We assume that the x87 and the SSE unit agree on the
-	 * rounding mode.  Reading the control word on the x87 turns
-	 * out to be about 5 times faster than reading it on the SSE
-	 * unit on an Opteron 244.
-	 */
-	__fnstcw(&__control);
-	return (__control & _ROUND_MASK);
-}
-
-static __inline int
-fesetround(int __round)
-{
-	int __mxcsr, __control;
-
-	if (__round & ~_ROUND_MASK)
-		return (-1);
-
-	__fnstcw(&__control);
-	__control &= ~_ROUND_MASK;
-	__control |= __round;
-	__fldcw(__control);
-
-	if (__HAS_SSE()) {
-		__stmxcsr(&__mxcsr);
-		__mxcsr &= ~(_ROUND_MASK << _SSE_ROUND_SHIFT);
-		__mxcsr |= __round << _SSE_ROUND_SHIFT;
-		__ldmxcsr(__mxcsr);
-	}
-
-	return (0);
-}
-
+/* C99 floating-point environment functions */
 int fegetenv(fenv_t *__envp);
 int feholdexcept(fenv_t *__envp);
-
-static __inline int
-fesetenv(const fenv_t *__envp)
-{
-	fenv_t __env = *__envp;
-	int __mxcsr;
-
-	__mxcsr = __get_mxcsr(__env);
-	__set_mxcsr(__env, 0xffffffff);
-	__fldenv(__env);
-	if (__HAS_SSE())
-		__ldmxcsr(__mxcsr);
-	return (0);
-}
-
+int fesetenv(const fenv_t *envp);
 int feupdateenv(const fenv_t *__envp);
 
 #if __BSD_VISIBLE
-
+/* Additional support functions to set/query floating point traps */
 int feenableexcept(int __mask);
 int fedisableexcept(int __mask);
-
-static __inline int
-fegetexcept(void)
-{
-	int __control;
-
-	/*
-	 * We assume that the masks for the x87 and the SSE unit are
-	 * the same.
-	 */
-	__fnstcw(&__control);
-	return (~__control & FE_ALL_EXCEPT);
-}
+int fegetexcept(void);
 
 #endif /* __BSD_VISIBLE */